Interview Workspace

This is your whiteboard. Write in §1–§9. Talk to the interviewer in chat. When a section is ready for the interviewer to look at, say so in chat — “requirements are up”, “done”, “look at §4”. They read only what the current phase covers.

§10 belongs to the interviewer. Do not write there. They will not write anywhere else.

Rough, ugly and fast beats neat and slow. Nobody scores your ASCII art. Half-sentences are fine; the interviewer is scoring your reasoning, and most of your reasoning happens in chat.

0 · Session Control

The interviewer fills this in at setup. Read it; don’t edit it — except to add assumptions you’ve been granted mid-round.


Question	(bank ID + title)
Company / level	(e.g. Google · `staff`)
Kind / domain
Attempt	(N on this question)
Mode / pace / difficulty
Started
Budget

The prompt (verbatim):

(the interviewer writes it here after reading it aloud)

Established facts — the interviewer’s answers to your clarifying questions:

You asked	Answer given

1 · Requirements & Scope

Budget: ~12 min. Signal the gate when §1.1–§1.4 are populated.

1.1 Functional requirements

What the system must do. Verbs. Number them so you can refer to them later.

1. 2. 3.

1.2 Non-functional requirements

Attach a number to every line you can. A non-functional requirement without a number is a wish.

Property	Target	Why this number
Scale (users / QPS / events/s)
Latency (p50 / p99)
Availability
Consistency
Durability / retention
Cost ceiling
Compliance / residency

1.3 Explicitly out of scope

Do not leave this empty. Cutting scope out loud is one of the cheapest points in the round, and an empty box here is the single most common thing the interviewer writes down as a negative.

-

1.4 Assumptions

Everything you’re taking as true that nobody told you. Revisit these aloud whenever the interviewer changes the premise.

-

2 · Estimation

Budget: ~9 min. Orders of magnitude, not decimals.

Quantity	Working	Result

2.1 What these numbers mean for the design

This subsection is the entire point of the phase. A number that doesn’t change a decision is decoration. Write the implication next to each figure — “X/s means one node can’t absorb it, so…”

-

Number kit, if you need it: 1 day ≈ 10⁵ s · QPS × 10⁵ ≈ per-day volume · 1 KB × 1M/s = 1 GB/s · DC round trip ≈ 0.5 ms · cross-region RTT ≈ 70–150 ms · SSD read ≈ 100 µs · memory read ≈ 100 ns · columnar compression ≈ 6× · 99.9 % ≈ 8.8 h/yr, 99.99 % ≈ 53 min/yr. Full kit: guides/playbook.md §11.

3 · API & Data Model

Budget: part of the ~18 min design phase. Skipping this is how a design gets a vague middle.

3.1 Core API / contracts

POST   /...
GET    /...

3.2 Entities & keys

Entity	Key	Fields that matter	Store

Partition key choice and why:

4 · High-Level Design

Budget: ~18 min including §3. Few boxes. Label every arrow with its protocol. Then be ready to walk one request end to end — you will be asked, every time.

       ┌──────────┐
       │  client  │
       └────┬─────┘
            │ HTTPS
            ▼

4.1 End-to-end happy path

One request or event, client to response, numbered steps. Write it — narrating it from a diagram you haven’t traced is where designs fall apart.

1. 2. 3.

4.2 What’s v1 and what’s later

Component	v1?	Why / trigger to add it

5 · Deep Dives

Budget: ~22 min. The interviewer drives this phase and picks the targets. This is where most of your score is decided.

5.1 Deep dive: (component)

The problem:

Options considered:

Option	Pros	Cons

Chosen, and why for this problem:

5.2 Deep dive: (component)

5.3 Curveball

A new requirement or failure lands here mid-phase. What changed, what didn’t, and what it cost.

6 · Failure Modes, Reliability & Operations

Budget: part of the ~9 min wrap. This is the manager bar — for an EM loop it carries as much weight as the architecture. Don’t leave it to the last two minutes.

What fails	Blast radius	Detection	Mitigation	Recovery

6.1 SLOs and alerting

SLI	SLO	Alert on	Page or ticket?

6.2 What breaks first as this grows

6.3 On-call reality

What wakes someone at 3 a.m., what the runbook says, and what you’d automate first.

7 · Cost

Rough monthly shape, and the biggest line item. “I don’t know” is a worse answer than a defensible order of magnitude.

Line item	Rough driver	Rough monthly

Biggest cost, and the lever that moves it:

8 · Summary — four sentences

You will be asked for this at the end, under time pressure. Write it here when you get there.

1. 2. 3. 4.

What I traded away, and what I’m least comfortable with:

9 · Scratch

Anything. Dead ends, arithmetic, half-thoughts. Nothing here is scored — but the interviewer may read it, and a visible discarded option with a reason attached is a positive signal, not a negative.

10 · Interviewer Notes — DO NOT EDIT

Appended live by the interviewer. Timestamps come from the shell, not from guesswork.

HH:MM:SS  elapsed  event